ERA (ERA Replicating Agents)

ERA (ERA Replicating Agents)

What we built

ERA is a Deno-native meta-agent that writes code to create other agents. It’s a self-improving AI meta-programming platform that continuously generates, tests, evaluates, and refines its own creations. Each agent can be promoted into a reusable “utility,” allowing ERA to grow its own capabilities over time.

Inspiration

Even as experienced developers, we found it surprisingly hard to get started building AI agents.
There are too many frameworks, protocols, and tools — and no clear starting point.
We built ERA to change that: a simple, self-improving system that lets anyone create custom agents to solve specific, real-world problems in their daily life.

What it does

ERA automates the full lifecycle of AI agent creation: Prompt → Generate → Test → Evaluate → Refine → Promote.

It includes both a CLI tool and a web interface for creating, managing, and refining agents. The system’s “FBI Director” orchestrator reviews each agent’s code, improves prompts, runs tests, and iterates automatically until success. Each attempt is tracked, versioned, and traced for transparency, allowing measurable improvement with every run.

Why It’s Useful

Rapidly prototype and evolve specialized AI agents
Automate debugging and code refinement
Build reusable utilities that improve future generations
Deploy instantly via Deno Deploy
Create a transparent feedback loop for AI development

In essence, ERA transforms agent creation into a self-sustaining, self-improving process.

How we built it

Untitled diagram-2025-10-12-181340

ERA is built entirely in Deno and TypeScript using a modular architecture centered on the FBI Director Orchestrator.

Component	Purpose
FBI Director	Core orchestrator that refines prompts, evaluates results, and drives iterative improvement
CLI Tool	Interactive command-line interface for agent creation and continuation
Hono Web Server + Alpine.js UI	Lightweight web interface for managing and monitoring agents
Tailwind CSS	Utility-first styling for a clean, minimal frontend
Daytona Sandbox	Executes and validates generated code safely
Wandb AI + Weave Tracing	Tracks inference runs, observability, and performance metrics
Groq / OpenAI / Zoom RTMS / Salesforce MCP	Optional inference and enterprise integrations
Deno Deploy	One-command production deployment
Val.town / Smallweb	Alternative lightweight hosting targets

Each module contributes to ERA’s continuous self-improvement loop and end-to-end observability.

What we built with

Tool / Protocol	How It Was Used
Deno	Core runtime environment for CLI, web server, and backend orchestration
TypeScript	Strongly-typed foundation for modular, scalable code
Hono	Fast Deno web framework for ERA’s API and interface
Alpine.js	Reactive frontend framework for lightweight interactivity
Tailwind CSS	Utility-based styling for the ERA dashboard
AG-UI	Provides a clean, modern interface for agent management and visualization
Daytona	Executes and validates generated agents inside secure sandboxes
Wandb	Handles AI inference calls and experiment tracking
Weave	Provides tracing and observability for every AI operation
Groq	Low-latency backend for LLM inference
OpenAI	Alternative inference engine for code generation and prompt refinement
Browserbase	Enables browser-based agent execution and web automation
Mastra	Framework for multi-agent workflows and orchestration demos
Tavily	Provides web search capabilities for research-driven agents
Zoom RTMS	Adds real-time messaging capability for collaborative agents
Salesforce MCP	Connects agent workflows to enterprise data and CRM systems
Deno Deploy	Seamless cloud deployment for both web and CLI layers
Val.town / Smallweb	Demonstrates serverless portability and compatibility